Designing a up Architecture That Actually Works W. Curtis Preston President/CEO The Storage Group What will we cover? What are the design options? -based, -free, Client-free, Server-free NDMP Using disk in your backup system What should I do with them? Sizing your server What up Software Are You Using? A. up Exec B. up Express C. BrightStor Enterprise up D. BrightStor ArcServe E. Commvault/Galaxy F. Omni/Data Protector G. Netup H. NetWorker I. NetVault J. TSM K. Other
Data General IBM IBM How Satisfied Are You With Your up Software? A. It s rocks. I d buy stock if I could. B. I d change some things if I could. C. These guys better start getting a clue. D. What were those other products again? What Are the Design Options? SAN: -free, Client-free, and Server-free backup NAS: NDMP filer to self, filer to filer, filer to server, & server to filer
IBM IBM IBM d i g i t a l Data General HEWLETT PACKARD -based ups Standard method Central backup server with network clients backing up across the Simplest, least expensive design -free ups How does this work? SCSI Reserve/Release Third-party queuing system Levels of drive sharing Restores Disk FC SCSI SCSI Disk FC FC SCSI SCSI Disk FC FC Switch SCSI/FC Router Client-free ups A 2c 2b Data 1 2a primary 3 disk set Transaction Logs backup mirror B up transaction logs to disk 2 A Data primary 1 disk set Transaction Logs backup m irror Establish backup mirror 4a 4b A Data 2 primary disk set Transaction Logs backup mirror 3 Split backup mirror and back it up 1
Client-free Restores 1a A 2a primary backup Data disk set mirror 2b Transaction Logs 2c Restore backup mirror from tape 1b Restore primary mirror from backup mirror Replay transaction logs from disk Server-free ups backup 2 primary mirror Data disk set or snapshot A Transaction Logs 3 SAN w/xcopy support Server directs client to take a copy-on-write snapshot Client and server record block and file associations Server sends XCOPY request to SAN 1 FileA Block A Block C Block D Virtual Disk Block B Provided by Disk Array FileB Block E Block F Server-less Restores Changing block locations Image-level restores File-level restores
ing up a Filer: NDMP Filer to self NDMP tape library Filer to filer Filer to server Server to Filer Filer to Server Filer to library Filer to Filer Server to filer Similar to server-free backups Other Server up Server Filer Filer Filer Filer to Self library library library Using NDMP Level of functionality depends on the DMA and filer vendors Robotic support Filer-to- support Filer-to-Server support Direct access restore support Image-level backup If You Have Filers, How Are They ed up? A. NDMP w/local tape drives on most filers B. NDMP filer-to-filer or filer-to-server C. up client running on the filer (usually a Windows or Linux filer) D. NFS/CIFS mount to a backup client/server E. Other
Disk to Disk to (D2D2T) ATA Arrays as low as $3/GB. Use ATA as target for all incremental backups. (Full, too, if you can afford it.) For offsite storage, duplicate all disk-based backups to tape. More information & ideas available at my other talk: Integrating Disk Into Your up System Sizing the up System Give it Enough Power Not enough tape drives drives that aren t fast enough Not enough slots in the tape library Not enough bandwidth to the server
Don t Give it Too Much Power Streaming tape drives must be streamed. If you don t, you will wear out your tape drives and decrease aggregate performance. Must match the speed of the pipe to the speed of the tape You can actually increase your throughput by using fewer tape drives. Server Size/Power I/O performance more important than CPU power CPU, memory, I/O expandability paramount Avoid overbuying by testing prospective server under load. Catalog/database Size Determine number of files (n) Determine number of days in cycle (d) (A cycle is a full backup and its associated incremental backups.) Determine daily incremental size (i = n *.02) Determine number of cycles on-line (c) 150-250 bytes per file, per backup Use a 1.5 multiplier for growth and error Index Size = (n + (i*d)) * c * 250 * 1.5
What is the predominant drive for open systems backup in your data center today? A. AIT B. SDLT C. LTO D. SAIT E. STK 9x40 F. IBM 3xx0 G. DDS H. VXA I. Other If you could start from scratch, what would you buy tomorrow? A. AIT B. SDLT C. LTO D. SAIT E. STK 9x40 F. IBM 3xx0 G. DDS H. VXA I. Other Number of Drives All -based up Buy twice as many backup drives as your network will support. Use only as many drives as the network will support. (You will get more with less.) Use the other half of the drives for duplicating.
Number of Drives Disk/ Combo -based up Buy disk system large enough to satisfy entire onsite retention period without deletion Buy enough tape drives to duplicate each night s backups. Duplicate each night s backups to tape, then take them out and send them offsite. should be large enough to hold three to four days of backups. (Only needs to hold duplicated tapes until they re sent offsite.) Number of Drives -Free backup Most large servers have enough I/O bandwidth to back themselves up within a reasonable time. Usually a simple matter of mathematics: 8 hr window, 8 TBs = 1 TB/hr = 277 MB/s 30 10 Mb/s drives, 15 20 MB/s drives Must have sufficient bandwidth to tape drives Filesystem vs. raw recoveries Allow drives and time for duplicating. Size - slots (all-tape environment) Should hold all onsite tapes Onsite tapes automatically expire and get reused. Only offsite tapes require physical management. Should automatically monitor library to ensure that each pool has enough free tapes before you go home. Watch for those downed drive messages.
Size -slots (disk/tape environment) D2D2t Do all backups to disk Disk big enough to hold all on-site copies Disk-based backups automatically expire and space gets reused needs to hold the latest set of copies (three or four days worth). Only off-site tapes require phys. mgmt. Should automatically monitor library and disk to ensure that each pool has enough free space before you go home Watch for those downed tape and disk drive messages D2d2T Do some backups to disk big enough to hold all on-site copies Disk-based backups automatically expire after copied to tape Disk needs to hold last night s backups before they go to tape Only off-site tapes require phys. mgmt. Should automatically monitor library and disk to ensure that each pool has enough free space before you go home Watch for those downed tape and disk drive messages D2D2t Do all backups to disk Disk big enough to hold all on-site copies Disk-based backups automatically expire and space gets reused needs to hold the latest set of copies (three or four days worth). Only off-site tapes require phys. mgmt. Should automatically monitor library and disk to ensure that each pool has enough free space before you go home Watch for those downed tape and disk drive messages D2d2T Do some backups to disk big enough to hold all on-site copies Disk-based backups automatically expire after copied to tape Disk needs to hold last night s backups before they go to tape Only off-site tapes require phys. mgmt. Should automatically monitor library and disk to ensure that each pool has enough free space before you go home Watch for those downed tape and disk drive messages
Configuring Your Server up all drives. Make sure you are streaming your drives. Create an automated monitoring system. Establish standards wherever possible, and use them! Resources Resources Directories of products to help you build a better backup system http://www.storagemountain.com Send questions to: curtis@thestoragegroup.com
More Questions? Mr. Preston will be available for one-on-one questions at the Ask the Experts area of the Exhibit Hall this evening from 5:00-6:00 pm. Sponsored by CA